Fal.ai has introduced an on-demand preference fine-tuning feature for image models, allowing enterprise users to tailor AI models based on user preferences directly collected through their apps. The process involves recording user rankings between images for the same prompt via a REST API, storing this data in a private dataset, and using it to fine-tune AI models like SDXL and SD1.5 with a single click. This customization aims to enhance user satisfaction without compromising the fast inference speeds of fal.ai's engine. The system encourages continuous data collection to keep models aligned with user preferences, and although the training functionality is currently limited to enterprise customers, the company is working on expanding access.